OpenGradient calls its network "permissionless", meaning no need for permission, no approval from anyone. On the protocol front, that's true; there’s no committee reviewing requests to run your node, no waiting list, no organizational KYC. I thought that was the end of the story until I read the section in the documentation about hardware requirements.
For an inference node to be registered in the system and start receiving real requests, the hardware running that node must first pass the TEE enclave authentication that is recorded on-chain. Intel TDX or AMD SEV are features available only on certain specific CPU ranges, not on the cheapest consumer models on the market. Not all servers are valid.
OpenGradient makes a calculated trade-off: absolute openness is not the goal, sufficient openness to decentralize while ensuring quality verification is the goal. OpenGradient requires a TEE because it’s the foundation of the entire guarantee layer the network needs for inference results to be reliable; without TEE attestations, the promise of verifiable AI is just a statement. OpenGradient accepts this compromise transparently; the hardware requirements are clearly stated in the documentation, with no fine print. But this means that the term "permissionless" in OpenGradient has a narrower meaning than most people think when they hear that word in crypto: the door is open for everyone in theory, but only those who invest in the right hardware can really cross the threshold.
$OPG